Research On Named Entity Recognition And Extraction Of Entities And Relations In Chinese Medical Field

In recent years,the popularization of applications such as medical knowledge graph,medical intelligent question-and-answer system and clinical decision support system has provided the basis for the improvement of medical personnel’s diagnosis and treatment efficiency,diagnosis and treatment ability as well as diagnosis and treatment quality,and is a hot area for the development of medical informatization construction.Named entity recognition and entity relationship extraction are two important subtasks for realizing medical knowledge graph.At present,the research on Chinese medical named entity recognition lacks the comprehensive utilization of Chinese character granularity and word granularity,which makes it difficult to improve the accuracy of Chinese medical named entity recognition results;in the research on Chinese medical entity relationship extraction,there are problems of entity extraction error accumulation and entity relationship triad overlap.In order to solve the above problems,the main research contents of this paper are as follows:First,in order to solve the problems of sparse granularity and specialized vocabulary in named entity recognition,this paper proposes a named entity recognition model based on gated recurrent units with the introduction of lexical information,and feature embedding and vector fusion for characters matching specialized words in the embedding layer;a new lexical gated unit is added in the context encoding layer,and the features required for entity recognition are automatically extracted using neural networks;the introduction of lexical information and a priori knowledge to achieve the improvement of Chinese medical named entity recognition.Secondly,to solve the problems of entity error accumulation and overlapping entities in entity relationship extraction,this paper proposes a relationship-driven Chinese medical entity relationship joint extraction model based on the attention mechanism.The model uses the attention mechanism to construct a specific sentence representation for each medical relationship;based on the named entity recognition model WI-NER,the head entity and tail entity corresponding to each relationship are extracted by sequence annotation;based on the influence of lexical information and context on the relationships,the entity extraction error accumulation problem and the relationship overlap problem are solved.Finally,experiments are conducted on three Chinese medical named entity recognition datasets to compare and analyze with other models,proving that the combined use of character granularity and word granularity units can effectively improve the accuracy of Chinese medical named entity recognition and validate the effectiveness of the named entity recognition model WI-NER proposed in this paper.Meanwhile,experiments are conducted on two Chinese medical entity relationship extraction datasets,and comparative analysis is performed with other models to solve the problem of entity extraction error accumulation,and the effectiveness of the entity relationship extraction model AM-ERE proposed in this paper is verified by analyzing the importance of each feature in entity relationship extraction.